openjev-sglang (Qwen3.6-35B-A3B on SGLang)
Data: JevBench v1.6.1, published 2026-10-06
Capability 58.7 · API offering, ranked on the API leaderboard · outside the Jev-class caps (cost 4.3× Jev (v1.5 reference); latency 2.1× Jev (v1.5 reference)), below the Capability ranking
Measured 2026-10-06 on A4 ∪ P, 600 items, equated (+2.57 I / +3.35 C); round score-a4-2.
no licence file in the repository as of 2026-09-19; Qwen3.6 weights keep their own terms
intelligence
38.8
calibration
78.6
speed
77.7
cost
35.6
Composite (secondary)
15.4 · Composite rank #21 of 27 on the API leaderboard
Cost and measurement conditions
$0.140 per 1,000 decisions (estimated)
v1.5.4 published cost per 1,000 decisions carried (pricing rules unchanged; v1.6 item lengths differ): ESTIMATE: exact base-model market reference (operator tariff excluded by 30-day rule)
p50 latency: 1.3 s
x2 (demo)
Breakdowns (not part of the score)
Chance-corrected competence (0 = chance, 100 = perfect) on A4 ∪ P (600 items: 300 open + 300 sealed; fewer sealed items than the full set, so wider uncertainty). Raw values, not equated. Category cells use A4+P+L1+L2+L3; per-type/tier cells use the original basis above. Radars against Jev.
| Request type | Open | Sealed |
|---|---|---|
| choice | 69.7 | 56.7 |
| noul | 32.4 | 3.0 |
| score | 30.9 | 24.5 |
Capability by subject topic
| Rules, policy & law | 23.1 | n=1133 |
|---|---|---|
| Coding & software | 53.1 | n=202 |
| Math & numbers | 0.0 | n=269 |
| Finance & commerce | 32.1 | n=178 |
| Support & operations | 35.7 | n=109 |
| Everyday language | 77.7 | n=130 |
| Safety & security | 49.0 | n=84 |
Use cases (TypeSafe categories)
| Model routing | 62.1 | n=267 |
|---|---|---|
| Legal & compliance | 25.1 | n=269 |
| Customer support | 39.2 | n=156 |
| E-commerce | 24.0 | n=88 |
| Insurance claims | 8.9 | n=92 |
| Risk assessment | 17.3 | n=88 |
| Financial crime | 9.3 | n=76 |
| Feature extraction | 14.3 | n=90 |
| Lead generation | 47.3 | n=86 |
| Recruiting | 17.6 | n=88 |
| Knowledge graphs | 8.6 | n=82 |
| LLM guardrails | 27.7 | n=82 |
| Moderation | 50.6 | n=69 |
| Code linting | 21.3 | n=81 |
| Search & retrieval | 68.0 | n=84 |
| Science | 4.6 | n=82 |
| Advertising | 14.9 | n=85 |
| Gaming | 0.0 | n=79 |
| Demand forecasting | 1.5 | n=72 |
Cells need at least 15 answered items; † fewer than 30 (indicative only, listed below the radar in the compare view).
Previous measurement (v1.5.x, not comparable)
Capability 70.8 · composite 29.0 · cost $0.140 per 1,000 decisions (estimated)
Last measured 2026-09-28, measurement release v1.5.0. Older method and scale; the current figures above replace it.